Kane CLI Online Hackathon
Build it with any agent. Verify it with Kane. Ship in three days.
A three-day online hackathon for developers who build with AI coding agents, and want to see what happens when you put a real verification layer next to them.
19 August to 21 August. $6,000 in prizes. Anywhere in the world. Hosted by TestMu AI.
Why this exists
AI coding agents have changed how software gets written. Features ship from prompts. Bugs get fixed in seconds. But one part of the loop never closed: when the agent ships something, someone still has to open a browser and click through to see if it actually works.
That's the bottleneck now. Not writing the code, trusting it.
Kane CLI was built for that gap. It's a plain-English browser automation tool that runs from your terminal. One command, no selectors, no framework. It opens a real local browser, runs the flow you describe, and returns a pass or fail with a video trace. It works as a developer utility. It also works as a tool that any AI coding agent can call directly.
That second part is the interesting one. An agent that can write code but can't check its own work is running open-loop. An agent that can call Kane can see what it just shipped, read the failure, and fix it, without a human in the middle.
We want to see what builders do when they close that loop. Which agent you use is entirely up to you.
The challenge
Build a working web app with an AI coding agent. Use Kane CLI to verify it works. Ship by 11:59 PM IST on 21 August.
That's the whole brief. The app can be anything you want: a tool, a dashboard, a side project you've been meaning to start, a game, an internal utility, a weird experiment. The only constraints are:
Any agent, any stack. Claude Code, Cursor, Copilot, Codex, Windsurf, Kiro, your own MCP setup, whatever you already build with. We don't care which one. We care that an agent is doing real work.
Kane CLI verifies the app. At minimum in your demo. The strongest submissions wire Kane into the agent directly: a hook that fires Kane on save, an agent that reads Kane's NDJSON output and iterates, a spec that compiles down to Kane flows.
It runs, it ships, and you can demo it on video.
You can work solo or in teams of up to four.
What "ready to ship" means
Your submission needs to clear three bars. Miss any of them and the judges won't be able to score the rest of what you built.
The app works end-to-end. A user can load it, complete the primary flow, and get a result. Not a screenshot, not a mock — the real thing, deployed or runnable locally with one command.
Kane CLI caught something or proved something. Show one of two things: "Kane caught this bug during the build, here's the failed run," or "Kane verifies these flows pass, here's the green run." Either counts. What doesn't count: Kane installed but never run, or one trivial flow tacked on at the end to qualify.
Your agent and Kane talked to each other. The agent's output triggered a Kane run, or a Kane result triggered the agent to do something. The cleanest version: a hook fires Kane on save, Kane fails, the agent reads the failure, edits the code, and the next save fires Kane again. Show us that moment. The tighter the loop, the higher the score.
How we'll judge it
Every submission is scored by a panel of TestMu AI engineers across four dimensions, weighted equally.
Ships: a working app with a real flow that runs end-to-end. Not slides.
Verified: Kane CLI actually exercised the app and caught or confirmed something meaningful.
Closed loop: agent built, Kane verified, result fed back to the agent. The tighter the integration, the higher the score.
Craft: did you reach for something interesting? Does this feel like a thing a developer would want to install tonight?
A polished todo app with one Kane flow tacked on at the end will lose to a weirder, scrappier thing where a hook fires Kane and the agent re-prompts itself based on what Kane finds. Both are valid submissions. The second is the one we're really hoping to see.
Three judges score every submission independently. Scores are averaged. Ties break on Closed loop.
Need an idea? Here are three lanes.
You don't have to pick a lane, anything that hits the brief is fair game. But if you're staring at a blank screen on day one, this is the menu.
Lane 1 - Apps that verify themselves
Build a web app where the agent ships the feature and Kane proves it works, without you opening the browser to check.
A todo app where every new feature comes with auto-generated Kane flows
A drag-and-drop dashboard builder that re-verifies every component after edits
A self-healing checkout flow where Kane catches a regression, the agent fixes it, and Kane re-verifies
A "prompt-to-feature" playground where users type "add a dark mode toggle" and watch the whole loop close in real time
An
npx create-verified-apptemplate that bakes Kane and an agent in from day oneA GitHub bot that drops into any AI-generated repo and adds Kane verification
A "QA-as-a-service" agent for non-technical founders — point it at a deployed app URL, get a bug report
The demo writes itself: open the app, trigger a change, watch the loop, show the green run.
Lane 2 - Verification baked into your workflow
Build the tool you wish existed in your day-to-day dev loop. Something that runs Kane automatically — on save, on push, on a schedule, on a Slack message — and uses an agent to decide what's worth your attention.
A
kane-watchfile watcher that re-verifies behavior on every saveA
gh kaneGitHub CLI extension that runs Kane on the current PR's preview deploy and posts results in the commentsAn MCP server that gives any agent a
verify_with_kanetoolA pre-commit hook that derives Kane flows from your git diff and blocks bad commits
Auto-bisect that walks back through commits to find the one that broke Kane
A visual canary that tells real regressions apart from harmless pixel drift
A conversion funnel watcher that runs your checkout every thirty minutes and pages on anomalies with the failed video attached
A doc-vs-product drift detector that runs the steps in your README and files an issue when they no longer work
The demo: install it on a real repo, or show the historical alerts the agent decided were worth paging on. Make the workflow getting tighter visible.
Lane 3 - Browser agents in the wild
Use Kane CLI as an agent's hands on the web. Not testing, the actual work the agent does for someone. This is where Kane stops being a QA tool and becomes browser infrastructure for any agent that needs to act on the web.
A job application autopilot that takes your resume and a job posting and submits the application, pausing on essay questions so you can answer them
A subscription killer that reads your bank statements and navigates each company's cancellation flow
A "renew everything" agent for domains, licenses, certifications, and memberships
A travel agent that books on real airline sites with no APIs in sight
Lead enrichment that visits each prospect's site and returns a one-pager
A Wikipedia speedrun bot
An agent playing a browser game with no API access — pure vision and clicks
A recursive ship-by-deadline where an agent writes a blog post about itself and uses Kane to publish it
The most useful versions solve a real, weird, specific problem. The most fun versions are pure spectacle. Both win.
Lane 4, Requirements that test themselves
Start from the PRD, not the app. Point Kane CLI at a spec, and it proposes use cases with every claim cited back to the doc, then designs ACs, scenarios, and one test per scenario as committable _test.md files. Run them, and the evidence pack shows what was proved versus what is still owed, coverage measured against ACs, not test counts.
A PRD-to-suite pipeline, all in the terminal:
kane-cli context ingest ./prd.md,kane-cli context extract, check what it captured withkane-cli context list, thenkane-cli design testsfor use cases, ACs, and scenarios as_test.mdfilesRun it with
kane-cli testmd run ./tests/<name>_test.mdand open the evidence pack: coverage against ACs, verdicts, and issues are all right thereA drift watcher: when the PRD changes,
kane-cli maintainreconciles what the suite now owes against the new requirementsA README auditor: ingest your own README, extract every claim it makes, and check whether the product still does those things
An agent that refuses to close a ticket until its acceptance criteria have a passing run in the evidence pack
A changelog verifier that ingests release notes and checks each shipped claim against production
The three days
All times Pacific, with India Standard Time alongside. The kickoff is recorded and posted within the hour if the live time doesn't work for you.
When & What:
Pre-hackathon details (sent over email)
Bonus tip: Attend Kane CLI sessions as part of the TestMu Conference 2026 to get insights.Kickoff stream (recorded & sent to you over email): 28th August, 9:30 PM IST (9:00 PM PT)
Building starts: 29th August-31 August
Submission deadline: 31 August, 11:59 PM IST
Judging: 31-2 September
Winner announcement live: on 3 September at 9:30 PM IST (9:00 AM PT) (30-minutes duration, with replay of top 3 demos)
Before you start
Two things, ideally done before the kickoff stream:
Install Kane CLI and run one flow. Run
npm install -g @testmuai/kane-cli, then try one of the examples from the docs. Ten minutes now saves you an hour on day one! A cheat sheet goes out with your confirmation email.Read the AI agent integration guide. It shows how to wire your agent to a Kane run. It's the fastest path to a closed loop.
You'll have free Kane CLI Pro access for the duration of the event, so don't worry about hitting limits.
If you attend the TestMu Conference 2026 the week before, several sessions are directly relevant prep. Recordings will be free to watch.
What to submit by 11:59 PM IST on 21 August
Four things, all through one form.
A GitHub repo: public, or invited access for the judges. Include a README with setup steps. The repo must be initialized after the kickoff stream.
A 3-minute demo video: screen recording, Loom or YouTube unlisted. Show the app working and show Kane running. Judges watch this first, so lead with the interesting part.
One paragraph: what you built, who it's for, which agent you used, and what Kane is doing in the flow.
A live URL or runnable command: judges should be able to see the app working in under 30 seconds.
Prizes
$6,000 USD in cash, split across 3 winners.
1st Place — $3,000
2nd Place — $2,000
3rd Place — $1,000
Participation Certificate to every team that submits a demo/entry before the deadline.
In addition to the cash, every winning team receives:
A 1:1 with the TestMu AI founders
Design partner status with input on the Kane CLI roadmap
3 months of Kane CLI Pro
A featured post on the TestMu blog
A showcase feature on testmuai.com/kane-cli/showcase
Amplification across TestMu channels
Verified with Kane CLI Badge: every team that ships a working project by the deadline gets the Verified with Kane CLI. Show up, ship, get the badge.
Who can enter
Anyone 18 or older
Solo or teams of up to four (one team captain receives the prize)
Open to builders anywhere in the world
No application, no curation, no cap, register and build
Rules
The app and the Kane CLI integration must be built during the event. Your repo must be initialized after the kickoff stream on 19 August, and we check commit history.
Use any AI coding agent, any framework, any stack. Kane CLI is the only required dependency.
If your project depends on a login or a paid service, hand the judges working credentials. We can't score what we can't run.
Submissions lock at the deadline. You can't push fixes after it.
If your project depends on a third-party service, include a fallback, a recorded run or a backup deploy, so judges can still see what you built.
One submission per team. If you want to build two things, build them with two teams.
Code stays yours. By submitting, you grant TestMu AI permission to feature your project in marketing materials with attribution.
The judges' decision is final.
What's included
Free Kane CLI Pro access for the duration of the event
A direct line to the engineers who built it, twice a day
A Slack channel full of builders working on the same problem
Early access to a tool that's about to be a much bigger deal
FAQs
Which agent do I have to use? None in particular. Claude Code, Cursor, Copilot, Codex, Windsurf, Kiro, an MCP setup you wrote yourself, pick the one you're fastest in. Kane CLI is the only required dependency. What we're scoring is how tightly you close the loop between whatever agent you chose and Kane.
Do I need a paid Kane CLI plan to enter? No. Every registrant gets free Kane CLI Pro access for the duration of the event. Winners get an additional three months on top.
Is this part of the TestMu Conference? It's hosted by the same team & we'll be talking about it all through the conference, but it runs the weekend after, 19 to 21 August. You don't need to have attended the conference to enter.
Can I bring a project I've already started? No. Your repo needs to be initialized after the kickoff stream on 19 August. Building on an idea you've been sitting on is fine, that's most good hackathon projects. Pushing a codebase you wrote in July is not.
Can I work solo? Yes. Solo entries are welcome. So are teams of up to four.
Do I need to know Playwright or Selenium? No. Kane CLI is plain English — you describe the flow, Kane runs it. If you do know Playwright, the Playwright export feature is built in and useful for the writeup, but it isn't required.
Can I use Kane CLI alongside Playwright or Selenium? Yes. Kane sits next to existing test setups. Many strong submissions will use Kane for flows that never made it into a Playwright suite , the quick checks, the one-offs, the things that didn't justify framework setup.
Do the demos have to be live? No. Everything is a recorded video, which means no scheduling and no time zone problems. The top three get played during the winners stream.
What if my demo video is over 3 minutes? Judges stop watching at 3:00. Put the good part first.
Where do I ask questions during the event? Over Slack channel (revealing shortly).
When do I find out if I won? Winners will be notified via email
Resources
Docs and quickstart — testmuai.com/kane-cli
AI agent integration guide — testmuai.com/kane-cli/agents.md
Example projects — github.com/testmuai/kane-cli-examples
GitHub repo — github.com/LambdaTest/kane-cli
Discord — https://discord.com/invite/kdkNUDHj
Submission form — revealing shortly, stay tuned
Questions — kane-hackathon@testmuai.com
Plain English. Real browser. Real apps. Pass or fail. Ship in three days.
See you online!